Design and optimization of an Atkinson cycle engine with the Artificial Neural Network Method

The Atkinson cycle engines have larger expansion ratio, thus higher thermal efficiency, which are more suitable for the hybrid fuel-electric vehicles than the conventional Otto cycle engines. Larger expansion ratio in an Atkinson cycle engine can be realized by increasing the geometrical compression ratio. Late Intake Valve Closure (LIVC) strategy is adopted to reduce the effective compression ratio to avoid the knock. However, the LIVC operation would reduce the effective displacement of the engine hence decrease the power density. There is a tradeoff between the thermal efficiency and Widely Open Throttling (WOT) torque/power. Computation-efficient nonlinear models for the baseline engine were built based on the Artificial Neural Network (ANN) technique. The ANN models were trained and tested using the data computed by a precisely calibrated GT-Power engine simulation model. Interactive effects of the LIVC, geometrical compression ratio, spark timing and air-to-fuel ratio on the fuel economy, WOT torque, knock intensity and exhaust temperature were deeply investigated. Optimization of the geometrical compression ratio and operating parameters was conducted based on the optimum ANN models. The optimization objective is to maximize the fuel economy, under the restriction conditions of WOT torque reduction percentage, knock intensity, and exhaust temperature. The optimum geometrical compression ratio was finally determined as 12.5. Experimental results obtained from the actual engine tests have validated the excellent prediction accuracy of the ANN models. Significant fuel economy improvement, of 6–13% at most WOT operating conditions, is obtained for the Atkinson cycle engine with acceptable compromise in the WOT torque.